Apprentissage par renforcement

ECTS : 3

Description du contenu de l'enseignement :

• Models: Markov decision processes (MDP), multiarmed bandits and other models
• Planning: finite and infinite horizon problems, the value  function, Bellman equations, dynamic programming, value and policy  iteration
• Basic learning tools: Monte Carlo methods, stochastic approximation, temporal-difference learning, policy gradient
• Probabilistic and statistical tools for RL: Bayesian approach,  relative entropy and hypothesis testing, concentration inequalities
• Optimal exploration in multiarmed bandits: the explore vs exploit tradeoff, lower bounds, the UCB algorithm, Thompson sampling
• Extensions: Contextual bandits, optimal exploration for MDP

Compétence à acquérir :

Reinforcement Learning (RL) refers to scenarios where the  learning algorithm operates in closed-loop, simultaneously using past  data to adjust its decisions and taking actions that will influence  future observations. Algorithms based on RL concepts are now commonly  used in programmatic marketing on the web, robotics or in computer game  playing. All models for RL share a common concern that in order to  attain one's long-term optimality goals, it is necessary to reach a  proper balance between exploration (discovery of yet uncertain  behaviors) and exploitation (focusing on the actions that have produced  the most relevant results so far).

The methods used in RL draw ideas from control, statistics and  machine learning. This introductory course will provide the main  methodological building blocks of RL, focussing on probabilistic methods  in the case where both the set of possible actions and the state space  of the system are finite.

Bibliographie, lectures recommandées :

• Reinforcement Learning: An Introduction, Richard S. Sutton & Andrew G. Barto, Second Edition, MIT Press, 2018
• Bandit Algorithms, Tor Lattimore & Csaba Szepesvári, Cambridge University Press, 2020